Slavic languages notoriously have gender for every object. In russian "friend" have two variants, one for male gender, one for female. Same at least with Ukrainian and Serbian. So when I translate sentence to English (which only have one word for "friend") I subconsciously use default word + she/he to empathize gender.
It's annoying habit, cause when I say or refer to "user" for example, I always assume it's "he/him".
I recently saw a video on youtube where guy followed and replicated Harvard experiment, which involved scanning news articles with specially trained llm, to find the moment to "buy low - sell high".
TL:DR Both: original llm and script made by that youtuber, failed miserably. Apparently result of the study was that market basically unpredictable.
What doesn't at this point? All repos on github contaminated at this point. In couple of years border between code created by humans and code created by llm's will be basically non existent.
Yeah, recent update completely borked game on both linux and windows. Devs released a patch soon after that added "comparability mode for linux peasants", but I saw suggestion for windows users to also use it, cause it solves same issue.
On the bright side - all those educated work migrants like me won't ever consider US for relocation. So clearly "anti migrant policy" is working as intended.
For some reason when I use sddm shutdown sometimes just doesn't work. Using GDM makes display doesn't work (weird hdmi quirk with bitdepth). So instead I login via TTY and start my session manually. At least then I can shutdown my system. Usually.
I have no clue why that triggered people so much. Never stated that I have any problem with that, nor that I need any help solving it. I leave my apartment like, once per month and cause living in EU - there no need to "travel" further than opposite street to buy anything I may need.
On top of that, point was that it's issue for other people in general. Explain how to setup tailscale to your grandma, I dare you.
Context limit is not really a problem on local models. Qwen3.6 can do up to ~256k tokens, it's not that far from what things like cursor uses. grok in cursor have exactly 256k tokens for example. Also, you can use opencode with custom config, where you need to set trimming close to that number.
That being said, quality wise it still kinda shit and looses to paid models. I think you need something like GLM 5.2 to compete, which needs 228gb at lowest 1bit quant (not including context size, which can be up to 1m tokens, so you can round up requirements for RAM up to 256gb), so yeah, the only thing that limits you locally — the fact that "AI" companies bought all supply of focking ram and we can't afford any.
Not everyone have static IP and device that can work as server. But yeah, it's not like you can't find work around: something like old android phone or spare notebook with pinggy on top would do the trick. Still, point stands - as for now it's less convenient option.
Music services still pretty useful. I'm using deezer for example, it can stream lossless audio and kinda okay with "radio" mode. Used spotify before, but it infuriated me by suggesting same music I already heard for 20-50 times. It's like at some point it just traps you in your own bubble of 100-200 songs.
The alternative would be downloading thousands of albums in flac and uploading to multiple devices. Plus you need to add new music, which includes searching for that mentioned music. So yeah, nah, I'm okay with paying subscription for now. Until they eventually go to shit like video services did. Then we will need to come up with "popcorn time" alternative for music I guess.
Can't wait for cold fusion to replace regular nuclear reactors. Then we will be finally able to boil water like no one ever did before!